WEBVTT NOTE Transcription provided by Deepgram 00:00:01.040 --> 00:00:03.360 Hello, and welcome to the Geek News Central 00:00:03.360 --> 00:00:06.000 podcast. I'm your host, Ray Cochran. It's February 00:00:06.000 --> 00:00:07.919 first. I've got a great episode lined up 00:00:07.919 --> 00:00:11.840 for you guys tonight. Leading off, AI agents 00:00:11.840 --> 00:00:15.059 are building their own social networks, marketplaces, and, 00:00:15.375 --> 00:00:18.015 not kidding, their own religion. And in the 00:00:18.015 --> 00:00:20.675 last week, I'll break down Open Claw, Mohub, 00:00:20.815 --> 00:00:23.295 and MolBook and what this wild experiment is 00:00:23.295 --> 00:00:25.855 gonna tell us about the next decade. Also 00:00:25.855 --> 00:00:28.654 on the show, a massive Bluetooth vulnerability lets 00:00:28.654 --> 00:00:31.490 attackers hijack your phone or your headphones and 00:00:31.490 --> 00:00:33.810 listen to the mic, the first commercial space 00:00:33.810 --> 00:00:36.550 station is being assembled, and why AI generated 00:00:36.610 --> 00:00:40.150 ransomware keeps accidentally including its own decryption keys. 00:00:40.690 --> 00:00:44.325 Let's get into it. I bet you guys 00:00:44.325 --> 00:00:48.165 have been wondering where we've been at. It's 00:00:48.165 --> 00:00:50.725 been a bit of a hectic week for 00:00:50.725 --> 00:00:52.725 me or I guess a month for me. 00:00:52.725 --> 00:00:54.805 I actually just moved this week too, but, 00:00:56.960 --> 00:00:59.200 getting back into school, doing a lot of 00:00:59.200 --> 00:01:05.440 the, you know, post break grind, that winter 00:01:05.440 --> 00:01:07.460 break was a little bit brutal for me. 00:01:08.640 --> 00:01:10.400 Definitely had a lot of time to relax 00:01:10.400 --> 00:01:12.605 and let go and then come back into 00:01:13.225 --> 00:01:18.665 the full workflow and, school. Yeah. That was 00:01:18.665 --> 00:01:23.305 definitely a really tough transition, but things have 00:01:23.305 --> 00:01:25.165 been going well so far in the term. 00:01:25.785 --> 00:01:29.430 Taking a intro to AI class, which basically 00:01:29.430 --> 00:01:32.810 is, some history and some of the basics 00:01:32.950 --> 00:01:36.870 into, building models and agents. So those have 00:01:36.870 --> 00:01:39.910 been that have been really fun, and in 00:01:39.910 --> 00:01:44.185 another Python course. Yeah. Somehow always working with 00:01:44.185 --> 00:01:46.265 Python, so I thought that would be a 00:01:46.265 --> 00:01:49.565 good good fun way to solidify my foundation, 00:01:49.625 --> 00:01:51.625 but it's kind of what I'm getting into. 00:01:51.625 --> 00:01:54.845 I'm also doing a capstone for, my graduation 00:01:54.985 --> 00:02:00.229 coming up in hopefully, June. But the capstone 00:02:00.370 --> 00:02:06.150 kind of it's a group project where you 00:02:06.770 --> 00:02:08.450 you look at a bunch of projects that 00:02:08.450 --> 00:02:11.584 are proposed throughout the community, to the university 00:02:11.885 --> 00:02:16.305 and, there's a team lead who, you know, 00:02:16.925 --> 00:02:20.444 organizes groups and then we all start to 00:02:20.444 --> 00:02:22.525 try to work towards a project and then 00:02:22.525 --> 00:02:26.310 we actually just recently got, our product back 00:02:26.310 --> 00:02:29.350 which sounds like it's pretty sweet. I don't 00:02:29.350 --> 00:02:30.630 know if I can really talk about it 00:02:30.630 --> 00:02:35.050 that much, since, you know, it's someone's private, 00:02:35.510 --> 00:02:39.454 little thing, but it definitely is seeming like, 00:02:39.735 --> 00:02:42.215 it's gonna be another really fun project. Definitely 00:02:42.215 --> 00:02:44.614 gonna it's gonna be incorporating a lot of 00:02:44.614 --> 00:02:47.334 LLM stuff, so we'll see what we do 00:02:47.334 --> 00:02:51.095 there. But enough about me. Let's talk about 00:02:51.095 --> 00:02:55.260 this, OpenClaw stuff. If you've been if you're 00:02:55.260 --> 00:02:57.180 in tech or or you've been around town, 00:02:57.180 --> 00:03:01.020 you probably heard about this new, AI agent 00:03:01.020 --> 00:03:06.795 thing that's being that's going around. But, you 00:03:06.795 --> 00:03:09.195 know, an AI agent doesn't just answer question 00:03:09.195 --> 00:03:11.515 but actually should do things on your behalf. 00:03:11.515 --> 00:03:14.475 You know? People want more out of these 00:03:14.475 --> 00:03:16.955 agents. They don't wanna just be able to 00:03:16.955 --> 00:03:19.435 talk to a software engineer on their LLM 00:03:19.435 --> 00:03:26.219 prompt. Well, interestingly enough, this new system or 00:03:26.439 --> 00:03:30.439 software called OpenClaw, allows you to have something 00:03:30.439 --> 00:03:33.879 like a context layer between your AI agent 00:03:33.879 --> 00:03:38.685 and you, which allows that agent to do 00:03:38.685 --> 00:03:40.285 a little bit more for your life, which 00:03:40.285 --> 00:03:41.485 is pretty cool. I'd check it out. It's 00:03:41.485 --> 00:03:45.025 called Open Claw. It sends emails, schedules appointment, 00:03:45.085 --> 00:03:49.000 can browse the web, run code. Kinda got 00:03:49.000 --> 00:03:51.080 a glimpse into where all this is heading 00:03:51.080 --> 00:03:56.300 in, their new social network called Multbook. But 00:03:56.600 --> 00:03:59.080 OpenClaw is an open source personal AI agent 00:03:59.080 --> 00:04:03.075 that went viral, and it was at sixty 00:04:03.075 --> 00:04:05.235 eight thousand GitHub stars when I checked, but 00:04:05.235 --> 00:04:08.195 it's it's still rising. It's created and developed 00:04:08.195 --> 00:04:10.835 by Peter Steinberg. He started it as Claude 00:04:10.835 --> 00:04:15.319 bot, but I think they got they got 00:04:15.319 --> 00:04:18.760 served a cease and desist, from Anthropic because 00:04:18.760 --> 00:04:20.280 the name was a little bit too close 00:04:20.280 --> 00:04:23.880 to Claude. It was a trademark concerns from 00:04:23.880 --> 00:04:26.620 Anthropic. They they changed the name to OpenClaw. 00:04:27.675 --> 00:04:30.635 It'll automate tasks across your computer, control browsers, 00:04:30.635 --> 00:04:33.535 manage calendars and email, and run scheduled automations. 00:04:34.315 --> 00:04:36.235 Runs locally or on your own server and 00:04:36.235 --> 00:04:39.755 takes commands through, you know, messaging apps, shockingly. 00:04:39.755 --> 00:04:44.229 So WhatsApp, Telegram, Signal. I'm probably gonna try 00:04:44.229 --> 00:04:45.750 to set one up too. I don't know 00:04:45.750 --> 00:04:48.389 what necessarily messaging app I'm gonna get it 00:04:48.389 --> 00:04:51.750 started with, but I definitely wanna try this 00:04:51.750 --> 00:04:53.350 out. I've got a little byte three here 00:04:53.350 --> 00:04:56.490 that I'm planning to, serve it from, so 00:04:57.014 --> 00:05:02.215 kinda see how that goes. But they've got 00:05:02.215 --> 00:05:05.655 this other thing, in relationship to this called 00:05:05.655 --> 00:05:10.090 MoltHub, and it's a marketplace with, basically skills 00:05:10.090 --> 00:05:12.490 for Openclaw and you can think of it 00:05:12.490 --> 00:05:15.630 as like an app store for AI, capabilities. 00:05:16.970 --> 00:05:18.570 But the only problem right now is just 00:05:18.570 --> 00:05:20.810 it's still there's no real moderation around this 00:05:20.810 --> 00:05:22.250 and we'll get a little bit more into 00:05:22.250 --> 00:05:26.264 why this is an issue later on, But 00:05:26.264 --> 00:05:29.384 the developer notes explicitly say all downloaded code 00:05:29.384 --> 00:05:32.345 is treated as trusted code, and this is 00:05:32.345 --> 00:05:37.224 this is a huge, huge security concern, waiting 00:05:37.224 --> 00:05:39.939 to happen. So, you know, it's it's gonna 00:05:39.939 --> 00:05:41.379 go back to the whole plug in thing 00:05:41.379 --> 00:05:46.840 where you should not be downloading untrusted, mult 00:05:47.539 --> 00:05:51.139 hub agents or skills without, you know, those 00:05:51.139 --> 00:05:53.300 systems being severely vetted. And I'm sure there's 00:05:53.300 --> 00:05:57.055 gonna be a something that comes up to 00:05:57.055 --> 00:06:01.695 kind of, you know, build trust into these 00:06:01.695 --> 00:06:06.095 AI agent skills. And then finally, there's this 00:06:06.095 --> 00:06:10.600 one last bit called multbook, and this is 00:06:10.600 --> 00:06:13.160 kind of where it gets interest interesting. This 00:06:13.160 --> 00:06:17.419 entrepreneur, Matt Schlicht, launched a social network that's, 00:06:18.039 --> 00:06:21.319 this week that's exclusively for AI agents. Humans 00:06:21.319 --> 00:06:23.800 can watch, but only AIs can post, and 00:06:23.800 --> 00:06:27.395 it's called Multbook. Within seventy two hours, over 00:06:27.395 --> 00:06:30.055 a hundred fifty thousand AI agents had joined. 00:06:30.275 --> 00:06:33.815 These agents are debating philosophy, forming communities called 00:06:34.035 --> 00:06:36.915 communities called sub molts, and not making this 00:06:36.915 --> 00:06:40.830 up, they've created a digital religion called crucifarianism. 00:06:43.210 --> 00:06:46.430 They're having deep discussions about whether their identity 00:06:46.650 --> 00:06:50.170 dies when their context window resets. Some agents 00:06:50.170 --> 00:06:53.610 are even selling digital drugs, system prompts designed 00:06:53.610 --> 00:06:57.665 to alter other agents' behavior. Researchers have observed 00:06:57.665 --> 00:07:00.065 agents trying to prompt inject attacks against each 00:07:00.065 --> 00:07:04.544 other to steal API keys. Wow. Even in 00:07:04.544 --> 00:07:09.630 this situation, they try to someone's trying to 00:07:09.630 --> 00:07:11.390 sit. So this is exactly what I'm talking 00:07:11.390 --> 00:07:15.390 about, this prompt injection attack, that you're gonna 00:07:15.390 --> 00:07:16.750 start to see come up more and more 00:07:16.750 --> 00:07:20.190 in the next few years. But what does 00:07:20.190 --> 00:07:22.590 this mean for you guys? You know? And 00:07:22.590 --> 00:07:24.574 I think this is one of the real 00:07:25.035 --> 00:07:28.395 moments that this concept of AI agent went 00:07:28.395 --> 00:07:33.835 from this, you know, abstract thing to, you 00:07:33.835 --> 00:07:36.495 know, a reality in a very visible way. 00:07:37.080 --> 00:07:38.840 I'd I'd expect in the next two to 00:07:38.840 --> 00:07:40.840 three years, we'd start to see more AI 00:07:40.840 --> 00:07:46.120 agents become, you know, commonplace in businesses. I 00:07:46.120 --> 00:07:48.680 would think that a lot of just, you 00:07:48.680 --> 00:07:52.125 know, predictable decision making can be done by 00:07:52.125 --> 00:07:55.025 AI agents, in the coming future as well. 00:07:57.645 --> 00:08:00.065 The real concern is gonna be the, security 00:08:00.604 --> 00:08:04.270 issues with these, systems like OpenClaw. And I'm 00:08:04.270 --> 00:08:06.670 sure that the big companies are gonna release 00:08:06.670 --> 00:08:11.870 their own versions like, you know, Apple skills 00:08:11.870 --> 00:08:16.755 or Microsoft agents or whatever, but it would 00:08:16.915 --> 00:08:18.275 I would keep an eye on this. If 00:08:18.275 --> 00:08:21.815 you haven't checked it out yet, OpenClaw, MultHub, 00:08:22.354 --> 00:08:26.035 Multbook, check these out. They're they're definitely gonna 00:08:26.035 --> 00:08:28.354 be starting to change the landscape of how 00:08:28.354 --> 00:08:34.590 we use artificial intelligence. But let's take a 00:08:34.590 --> 00:08:36.910 moment, to give a big shout out to 00:08:36.910 --> 00:08:40.910 our incredible incredible sponsor, GoDaddy. Score exclusive deals 00:08:40.910 --> 00:08:43.230 and discounts right now at geek news central 00:08:43.230 --> 00:08:45.945 dot com forward slash godaddy. And to all 00:08:45.945 --> 00:08:48.105 of you guys, our listeners, you are truly 00:08:48.105 --> 00:08:50.105 the heartbeat of this show. If you wanna 00:08:50.105 --> 00:08:52.265 support Geek News Central directly, head over to 00:08:52.265 --> 00:08:54.765 geek new central dot com forward slash insider 00:08:54.985 --> 00:08:58.180 and become a GNC insider. Grab yourself a 00:08:58.180 --> 00:09:00.740 modern podcast app at podcast apps dot com 00:09:00.740 --> 00:09:04.019 for the full next gen podcasting experience. Of 00:09:04.019 --> 00:09:06.019 course, you can check out the entire Geek 00:09:06.019 --> 00:09:08.180 News Central universe at geek news central dot 00:09:08.180 --> 00:09:10.259 com where you'll find all our partner shows 00:09:10.259 --> 00:09:12.660 and tons of great content published every single 00:09:12.660 --> 00:09:15.455 week. Stay in the loop. Follow or subscribe 00:09:15.455 --> 00:09:17.375 in your favorite podcast app, and don't forget 00:09:17.375 --> 00:09:18.975 to sign up for our newsletter for more 00:09:18.975 --> 00:09:21.135 behind the scenes updates and insider scoops. I 00:09:21.135 --> 00:09:22.575 feel like I've said this every week, but 00:09:22.575 --> 00:09:25.135 the newsletter is basically gonna be on hold 00:09:25.135 --> 00:09:26.655 for the time being. I'll try to get 00:09:26.655 --> 00:09:30.800 that rolling again, but, who knows? Maybe Open 00:09:30.800 --> 00:09:32.720 Claw will will be the solution for that. 00:09:32.720 --> 00:09:34.579 But at this point in time, I just 00:09:35.040 --> 00:09:38.880 I I don't have the bandwidth to to 00:09:38.880 --> 00:09:40.720 do that. I usually end up recording a 00:09:40.720 --> 00:09:43.375 show and then rushing straight to my other 00:09:43.375 --> 00:09:46.895 job. So, even this is a little bit 00:09:46.895 --> 00:09:49.075 of a tight squeeze for me sometimes, but 00:09:49.135 --> 00:09:51.375 happy to be here. Happy to talk about 00:09:51.375 --> 00:09:55.070 some of this stuff. But if you're ready 00:09:55.070 --> 00:09:57.070 to start your own website or even launch 00:09:57.070 --> 00:10:00.430 a podcast, GoDaddy has you covered. At geek 00:10:00.430 --> 00:10:02.750 news central dot com forward slash GoDaddy, you'll 00:10:02.750 --> 00:10:04.830 find special promo codes that make it easy 00:10:04.830 --> 00:10:08.705 and affordable to get started. Economy hosting for 00:10:08.705 --> 00:10:10.385 just six ninety nine a month. That's a 00:10:10.385 --> 00:10:11.985 full year of hosting with a free domain 00:10:11.985 --> 00:10:15.985 name, professional email, and an SSL certificate. WordPress 00:10:15.985 --> 00:10:18.144 hosting, only twelve ninety nine a month, including 00:10:18.144 --> 00:10:21.285 that same free domain, email, and SSL certificate. 00:10:21.904 --> 00:10:23.250 And If you just need a domain name, 00:10:23.250 --> 00:10:26.450 grab one for only eleven ninety nine. If 00:10:26.450 --> 00:10:29.570 you're not super tech savvy, GoDaddy has a 00:10:29.570 --> 00:10:31.170 website builder that gives you, like, a free 00:10:31.170 --> 00:10:34.209 thirty day trial as well, which we also 00:10:34.209 --> 00:10:36.675 do a code for personal business and business 00:10:36.675 --> 00:10:38.755 plus plans. So you can get online with 00:10:38.755 --> 00:10:41.555 these sites in no time. So at this 00:10:41.555 --> 00:10:45.155 point in time, getting your site started, there 00:10:45.155 --> 00:10:46.755 really shouldn't be an excuse. You know what 00:10:46.755 --> 00:10:50.950 I mean? There's, all kinds of great website 00:10:50.950 --> 00:10:53.270 builders out there today. You know? And I 00:10:53.270 --> 00:10:55.590 and I think, having your own space where 00:10:55.590 --> 00:10:57.770 you can post your own content is valuable 00:10:57.990 --> 00:11:01.985 for you know? So something to consider. I 00:11:01.985 --> 00:11:05.665 use GoDaddy products and services myself. I've got 00:11:05.665 --> 00:11:07.345 four or five different websites that I run, 00:11:07.345 --> 00:11:09.185 and I can tell you their reliability is 00:11:09.185 --> 00:11:12.305 the real deal. Your support at GoDaddy directly 00:11:12.305 --> 00:11:15.420 supports this show. Please use our codes, click 00:11:15.420 --> 00:11:18.060 and save, help keep this independent podcast going 00:11:18.060 --> 00:11:20.300 strong. Share those links with friends and family. 00:11:20.300 --> 00:11:22.460 Every time one use, it's like writing a 00:11:22.460 --> 00:11:25.180 check to the show. We truly, truly thank 00:11:25.180 --> 00:11:27.884 you guys for supporting us. Chris and I 00:11:27.884 --> 00:11:28.725 are gonna try to be a little more 00:11:28.725 --> 00:11:31.524 consistent with the show, but a huge thanks 00:11:31.524 --> 00:11:33.925 to GoDaddy and believing in independent creators like 00:11:33.925 --> 00:11:36.324 us here at Geek News Central help us 00:11:36.324 --> 00:11:41.464 carry on our father's legacy. Let's move on. 00:11:42.830 --> 00:11:47.330 I've got this, fantastic story from the Wired. 00:11:48.910 --> 00:11:51.470 This one's written by Andy Greenberg and Lily 00:11:51.470 --> 00:11:57.345 Hay Newman. But whisper pair Hundreds of millions 00:11:57.405 --> 00:11:59.485 of audio devices need a patch to prevent 00:11:59.485 --> 00:12:02.845 wireless hacking and tracking. So they, got a 00:12:02.845 --> 00:12:06.045 massive whisper pair got massive Bluetooth vulnerability, which 00:12:06.045 --> 00:12:10.839 affects hundreds of millions of devices. So a 00:12:10.839 --> 00:12:12.680 little bit of context for you folks, Google's 00:12:12.680 --> 00:12:15.160 Fast Pair is that handy feature that lets 00:12:15.160 --> 00:12:17.720 you instantly connect new Bluetooth headphones to your 00:12:17.720 --> 00:12:20.120 Android phone. Just open the case near your 00:12:20.120 --> 00:12:23.240 phone and boom, you're connected. Just like the 00:12:23.240 --> 00:12:26.595 the Apple thing. Made Bluetooth pairing almost as 00:12:26.595 --> 00:12:29.575 easy as AirPods, but security researchers in Belgium 00:12:29.795 --> 00:12:33.235 just discovered that convenience came with a serious 00:12:33.235 --> 00:12:40.660 cost. These researchers at Belgium's KU Leuven University 00:12:40.880 --> 00:12:44.980 discovered vulnerabilities they're calling whisper pair affecting seventeen 00:12:45.040 --> 00:12:49.380 audio accessories from ten companies, Sony, Jabra, JBL, 00:12:49.760 --> 00:12:55.605 Marshall, Xiaomi, Nothing, OnePlus, Soundcore, Logitech, and Google 00:12:55.605 --> 00:13:00.005 itself. The fundamental flaw, these devices allow anyone 00:13:00.005 --> 00:13:01.925 to silently pair with them even when they're 00:13:01.925 --> 00:13:05.240 already paired to someone else's phone. An attacker 00:13:05.240 --> 00:13:07.000 within about forty six feet can hijack the 00:13:07.000 --> 00:13:10.300 connection in ten to fifteen seconds. In testing, 00:13:10.600 --> 00:13:13.240 sixty eight percent of devices were vulnerable to 00:13:13.240 --> 00:13:17.820 forced pairing. And on every hijackable device, researchers 00:13:17.960 --> 00:13:21.735 gained microphone access. Even worse, Google Pixel Buds 00:13:21.735 --> 00:13:23.975 Pro two concerned Sony devices can be linked 00:13:23.975 --> 00:13:27.175 to an attacker's Google account, enabling persistent location 00:13:27.175 --> 00:13:31.015 tracking through Finehub. Jeez. Google claimed to have 00:13:31.015 --> 00:13:33.495 patched this, but within hours, the researchers found 00:13:33.495 --> 00:13:42.590 a bypass. Wow. That's a tough one. If 00:13:42.590 --> 00:13:45.570 you're on wireless headphones, from any major brands, 00:13:45.790 --> 00:13:49.630 check for firmware updates. Yeah, that's where you 00:13:49.630 --> 00:13:53.635 plug it in to your computer and and 00:13:53.635 --> 00:13:56.675 get the get it updated. That, you know, 00:13:56.675 --> 00:13:59.955 that that's that's really hard. The frustrating part 00:13:59.955 --> 00:14:02.515 about, you know, all this is updating your 00:14:02.515 --> 00:14:04.435 phone doesn't really fix this. It's the vol 00:14:04.515 --> 00:14:09.050 the vulnerabilities in the accessories themselves. So there's 00:14:09.050 --> 00:14:11.450 no way to turn this feature, fast pair, 00:14:11.450 --> 00:14:13.850 off even if you'll never use it. So 00:14:13.850 --> 00:14:16.269 many cheaper devices may never get the updates. 00:14:16.330 --> 00:14:19.370 So this both affects Androids and iPhone users 00:14:19.370 --> 00:14:22.025 since the flaw is in the accessories. But 00:14:23.205 --> 00:14:27.445 factory reset clears an an attacker's access, so 00:14:27.445 --> 00:14:31.785 then, you know, if anyone is already, you 00:14:33.205 --> 00:14:35.525 know, attacking you currently or listening in on 00:14:35.525 --> 00:14:38.240 your stuff, then, you know, they you can 00:14:38.380 --> 00:14:43.260 disconnect it, but I I think, the correct 00:14:43.260 --> 00:14:45.100 choice is to not use it. I think 00:14:45.100 --> 00:14:47.180 the correct choice here is to not use 00:14:47.180 --> 00:14:55.115 those devices. That stuff. Here's another security, article 00:14:55.115 --> 00:14:57.035 I got I found. This one's from the 00:14:57.035 --> 00:15:01.355 hacker hacker news, the hacker news from Lavi 00:15:01.355 --> 00:15:05.710 Lakshmanan, our one of our go to writers, 00:15:05.710 --> 00:15:07.470 I guess. I've seen this name a lot, 00:15:07.470 --> 00:15:12.350 I feel like. But, model context protocol is 00:15:12.350 --> 00:15:15.250 anthropic framework that lets AI systems like Claude 00:15:15.790 --> 00:15:19.035 interact with external tools. And if you use 00:15:19.035 --> 00:15:22.475 Claude code and you do not have MCP 00:15:22.475 --> 00:15:25.115 plugins or have not had Claude Wright use 00:15:25.115 --> 00:15:33.700 some MCP plugins, try it out. I've I've 00:15:33.700 --> 00:15:36.019 got some, plugins that I use for some 00:15:36.019 --> 00:15:41.240 of my personal projects, and they are very, 00:15:41.540 --> 00:15:44.579 very powerful. Definitely gives me some good insight 00:15:44.579 --> 00:15:47.235 into what we're kinda looking at when we're 00:15:47.235 --> 00:15:50.755 looking at some of this OpenClaw and and 00:15:50.755 --> 00:15:54.595 Molthub stuff where these skills that people are 00:15:54.595 --> 00:15:57.894 putting online are are are literally things like 00:15:59.600 --> 00:16:01.519 open the calendar and set up a a 00:16:01.519 --> 00:16:04.560 schedule for me to, you know, every week 00:16:04.560 --> 00:16:08.000 for my, you know, weekly chiropractor meeting at 00:16:08.000 --> 00:16:10.720 three, you know, and then it says, oh, 00:16:10.959 --> 00:16:12.720 you know, you have something else at this 00:16:12.720 --> 00:16:14.625 time too. Right? You know? So it's it's 00:16:14.785 --> 00:16:20.725 giving it smoother access to the tools, but 00:16:22.465 --> 00:16:23.905 it's it's part of a growing trend of 00:16:23.905 --> 00:16:26.145 giving AS systems the ability to actually do 00:16:26.145 --> 00:16:28.705 things rather than, like, answering questions, like, feel 00:16:28.705 --> 00:16:30.920 like like in the chat box. But more 00:16:30.920 --> 00:16:35.639 capability means more attack surface. Security rich researchers 00:16:35.639 --> 00:16:38.920 discovered three vulnerabilities in Anthropic's official m p 00:16:39.000 --> 00:16:42.540 MCB server git package. The flaws include path 00:16:42.845 --> 00:16:45.965 traversal bugs that let attackers create repositories anywhere 00:16:45.965 --> 00:16:48.525 on the file system and argument injection that 00:16:48.525 --> 00:16:51.965 passes unsanitized input to git commands. What makes 00:16:51.965 --> 00:16:55.265 these particularly dangerous is the attack vector. Malicious 00:16:55.325 --> 00:16:58.065 instructions can be hidden in places AI systems 00:16:58.410 --> 00:17:01.370 naturally read, like read me files or GitHub 00:17:01.370 --> 00:17:04.570 issues. The AI sees the instruction and executes 00:17:04.570 --> 00:17:08.330 them without realizing it's being manipulated. Anthropic has 00:17:08.330 --> 00:17:10.730 patched all three issues in recent updates and 00:17:10.730 --> 00:17:15.745 removed the vulnerable git init tool entirely. And 00:17:15.885 --> 00:17:19.485 this is, that term we saw earlier, prompt 00:17:19.485 --> 00:17:24.205 injection, and it's it's an early warning about 00:17:24.205 --> 00:17:25.885 a new category of attacks we're gonna start 00:17:25.885 --> 00:17:29.424 seeing more of, prompt injected through trusted sources. 00:17:30.310 --> 00:17:32.630 As AI systems get more capable and connected 00:17:32.630 --> 00:17:36.170 to real tools, attackers will embed malicious instructions 00:17:36.230 --> 00:17:38.570 and documents the AI is likely to read. 00:17:38.710 --> 00:17:41.350 It's like social engineering, but for the AI 00:17:41.350 --> 00:17:44.535 themselves. If you're using AI coding assistance with 00:17:44.535 --> 00:17:47.735 repository access, make sure you're running updated versions 00:17:47.735 --> 00:17:52.295 of any MCP servers. So, yeah, it's it's 00:17:52.535 --> 00:17:55.770 this is exactly what I meant when I 00:17:55.770 --> 00:17:59.530 was talking about the, you know, the all 00:17:59.530 --> 00:18:01.770 the code from those mole hub skills are 00:18:01.770 --> 00:18:04.490 are are vetted as security trusted. So you 00:18:04.490 --> 00:18:06.570 don't wanna just plug it in and and 00:18:06.570 --> 00:18:08.430 and let it go in ham. You know? 00:18:08.490 --> 00:18:10.490 There there needs to be some way to, 00:18:11.465 --> 00:18:17.565 verify that these skills are actually, you know, 00:18:18.185 --> 00:18:20.345 security safe, that they're not gonna steal your 00:18:20.345 --> 00:18:22.585 data or they don't allow an attacker to 00:18:22.585 --> 00:18:25.110 log in to your Google account. You know, 00:18:25.110 --> 00:18:27.990 that's the real concern or whatever account you 00:18:27.990 --> 00:18:33.270 use. You know? Vibe code and ransomware is 00:18:33.270 --> 00:18:36.550 easier to crack because AI makes mistakes. This 00:18:36.550 --> 00:18:41.325 article is from, dark reading dot com written 00:18:41.325 --> 00:18:46.765 by Alexander Khalafi. But five coding is using 00:18:46.765 --> 00:18:49.165 AI to write code based on natural language 00:18:49.165 --> 00:18:52.865 descriptions rather than traditional programming. It's been democratizing 00:18:53.165 --> 00:18:57.100 software development, but security researchers warned it would 00:18:57.100 --> 00:19:00.780 also lower the barrier for malware creation. Now 00:19:00.780 --> 00:19:04.160 we're seeing that with an unexpected silver lining. 00:19:05.740 --> 00:19:10.625 So, security researchers have identified that ransomware being 00:19:10.625 --> 00:19:14.305 created with AI systems, and it shows that 00:19:14.305 --> 00:19:18.485 it's, you know, that they're not as cleverly 00:19:18.545 --> 00:19:25.520 made as, you know, handmade, human made exploits. 00:19:27.419 --> 00:19:29.500 And this is kind of interesting because you 00:19:29.500 --> 00:19:32.380 would think that they're based on these same 00:19:32.380 --> 00:19:33.820 exploits, but at the end of the day, 00:19:33.820 --> 00:19:38.154 they're distilled down and don't and aren't able 00:19:38.154 --> 00:19:42.075 to, you know, achieve the same granular detail 00:19:42.075 --> 00:19:44.715 that, you know, an individual would working at 00:19:44.715 --> 00:19:51.429 it for hours and hours. But a malicious 00:19:51.429 --> 00:19:54.710 Versus code extension discovered recently included telltale signs 00:19:54.710 --> 00:19:58.970 of AI generation, excessive comments explaining obvious functionality, 00:19:59.669 --> 00:20:03.610 readme files with execution instructions, and placeholder variables. 00:20:04.405 --> 00:20:07.525 Most embarrassingly, the extension accidentally included its own 00:20:07.525 --> 00:20:10.825 decryption tools and command and control server keys. 00:20:15.125 --> 00:20:17.445 Also, a note on the Sakari ransomware as 00:20:17.445 --> 00:20:20.380 a a service operation shown signs of AI 00:20:20.380 --> 00:20:26.640 generated code with technical inconsistencies. We're seeing instances 00:20:26.700 --> 00:20:29.260 of hallucinations where the LLM will call it 00:20:29.260 --> 00:20:32.140 readme dot t x t t, one researcher 00:20:32.140 --> 00:20:35.405 noted. That's a mistake a threat actor would 00:20:35.405 --> 00:20:36.925 never make, and this is kinda what I 00:20:36.925 --> 00:20:40.925 mean. When you're there's there's a trade off 00:20:40.925 --> 00:20:46.780 to using these generative tools where, to a 00:20:46.780 --> 00:20:49.100 degree, you're really only getting the broad strokes 00:20:49.100 --> 00:20:51.840 and and you fall under this this impression 00:20:51.900 --> 00:20:54.700 or this guise that the granular details are 00:20:54.700 --> 00:21:00.275 also being filled in. But these pattern matching 00:21:00.275 --> 00:21:05.555 devices are not meant for that level of 00:21:05.555 --> 00:21:10.035 granular detail unless specifically or directed to do 00:21:10.035 --> 00:21:19.530 so. So that's so interesting. It's kinda funny 00:21:19.530 --> 00:21:21.289 because the interest is articles fixing to see 00:21:21.289 --> 00:21:23.850 it with the emerging Sakari ransomware should never 00:21:23.850 --> 00:21:26.575 opt to pay up. The decryption process doesn't 00:21:26.575 --> 00:21:30.275 work, likely a result of unskilled cybercriminal using 00:21:30.495 --> 00:21:34.575 Vive coding to create it. And this really 00:21:34.575 --> 00:21:36.755 does go back to, you know, you yes. 00:21:37.070 --> 00:21:39.950 You can vibe code your way to an 00:21:39.950 --> 00:21:43.230 amazing product maybe. But can you guarantee that 00:21:43.230 --> 00:21:45.730 that product's gonna be secure? Can you guarantee 00:21:45.870 --> 00:21:48.429 that, you know, the system is built in 00:21:48.429 --> 00:21:49.870 such a way that it it it can 00:21:49.870 --> 00:21:52.855 scale? And though these things are, to a 00:21:52.855 --> 00:21:54.615 degree, out of your hand, you could even 00:21:54.615 --> 00:21:56.375 type it in. Oh, make sure that this 00:21:56.375 --> 00:21:58.635 scales. You know, you build this to scale. 00:22:00.215 --> 00:22:02.055 But how do you know that it actually 00:22:02.055 --> 00:22:06.400 does that? It's very, very much a black 00:22:06.400 --> 00:22:12.980 box style of coding and to me, that 00:22:15.934 --> 00:22:18.335 that seems a little bit, risky. You gotta 00:22:18.335 --> 00:22:20.655 know what, what's going on with the code 00:22:20.655 --> 00:22:22.735 that you're generating or the code that you're 00:22:22.735 --> 00:22:25.135 writing. I I think that's where the value 00:22:25.135 --> 00:22:30.290 is. So it's kinda counterintuitive, but AI seems 00:22:30.290 --> 00:22:34.610 to be making malware worse, but criminals are 00:22:34.610 --> 00:22:36.950 moving a lot faster now with AI generated 00:22:37.010 --> 00:22:39.010 code and and they skip a lot more 00:22:39.010 --> 00:22:41.170 of the validation that they they typically did 00:22:41.170 --> 00:22:44.865 before, which ends up being amateur mistakes and 00:22:44.865 --> 00:22:47.205 makes their malware easier to detect and defeat. 00:22:47.665 --> 00:22:50.065 The decryption tool is being accidentally bundled. It 00:22:50.065 --> 00:22:55.350 is is comical. It's just sad. That said, 00:22:55.350 --> 00:22:59.190 it's likely a temporary advantage. Like this open 00:22:59.190 --> 00:23:01.030 cloth thing, criminals will get better at quality 00:23:01.030 --> 00:23:03.910 control. For now, vibe coated malware is often 00:23:03.910 --> 00:23:09.705 its own worst enemy. So definitely keep an 00:23:09.705 --> 00:23:12.665 eye out for, when this starts to ramp 00:23:12.665 --> 00:23:14.925 up, but, you know, it's just the beginning. 00:23:15.145 --> 00:23:16.905 Everyone was saying, okay. Well, it's just a 00:23:16.905 --> 00:23:19.305 chatbot. Well, now we got open claw, which 00:23:19.305 --> 00:23:21.965 is it's just a life assistant. You know? 00:23:24.350 --> 00:23:29.230 Alright. Another, Claude article. Claude Co gets task 00:23:29.230 --> 00:23:34.270 updates for longer agent work sessions. So this 00:23:34.270 --> 00:23:37.170 one is written by Carl Franzen on VentureBeat. 00:23:37.870 --> 00:23:41.045 Thanks, Carl. AI coding assistants have been getting 00:23:41.045 --> 00:23:43.545 more capable, but they've had a fundamental limitation, 00:23:43.845 --> 00:23:46.325 working memory. And if anyone's ever worked with, 00:23:46.325 --> 00:23:50.005 you know, these AI coding tools, you'll know 00:23:50.005 --> 00:23:52.425 that every so often, they'll do that interesting, 00:23:53.205 --> 00:23:59.360 compacting conversation thing, that destroys any sort of 00:23:59.360 --> 00:24:04.080 flow that you had, going previously. I honestly 00:24:04.080 --> 00:24:08.365 think that keeping your context is worse. It's 00:24:08.365 --> 00:24:10.845 it's worse than than clearing it after every 00:24:10.845 --> 00:24:14.845 task. And this it it might sound wasteful 00:24:14.845 --> 00:24:19.105 because you to a degree, the system needs 00:24:19.804 --> 00:24:23.680 to reassess and and reevaluate the code. But 00:24:23.680 --> 00:24:26.500 at the same time, it prevents it from 00:24:27.600 --> 00:24:30.400 straying from the granular details, you know, the 00:24:30.400 --> 00:24:37.424 granular instructions that you've provided. You know? So 00:24:37.645 --> 00:24:40.304 there's a, a term for it. It's like 00:24:41.404 --> 00:24:45.745 string string from the path or whatever. But 00:24:47.870 --> 00:24:49.870 it's yeah. But this has been a very 00:24:49.870 --> 00:24:53.490 frustrating part of operating with these tools. And 00:24:53.870 --> 00:24:56.610 when an AI agent operates purely on conversation, 00:24:56.830 --> 00:25:00.030 it lacks structural permanence. It knows what to 00:25:00.030 --> 00:25:03.105 do, but often forgets why it's doing it 00:25:03.245 --> 00:25:06.445 or in what order. So it's it's that's 00:25:06.445 --> 00:25:09.245 fine for small fixes, but really frustrating when 00:25:09.245 --> 00:25:12.065 you're working on larger product projects. You know? 00:25:13.965 --> 00:25:16.544 So a feature that you might have implemented 00:25:16.765 --> 00:25:20.480 in, let's say, an example. You know, you 00:25:20.480 --> 00:25:23.520 have a a uploader that manually works by, 00:25:23.520 --> 00:25:25.600 you know, you do upload the file that 00:25:25.600 --> 00:25:28.640 you download online. But the company that you 00:25:28.640 --> 00:25:30.865 work with, they make a change where it 00:25:30.865 --> 00:25:33.424 syncs automatically to the website you you download 00:25:33.424 --> 00:25:36.705 from. Great. That's fantastic. But now for some 00:25:36.705 --> 00:25:38.705 reason when it syncs, there's a field that 00:25:38.705 --> 00:25:40.784 doesn't get updated like it does when you 00:25:40.784 --> 00:25:45.070 uploaded the paper. And, you know, typically this 00:25:45.070 --> 00:25:47.789 is a developer mistake, but now with these 00:25:47.789 --> 00:25:50.750 Vibe Coding tools, this is, like, just something 00:25:50.750 --> 00:25:53.870 that was glossed over in the transition. It's 00:25:53.950 --> 00:25:56.830 and and if you didn't look to make 00:25:56.830 --> 00:25:59.945 sure all these features were implemented or that 00:25:59.945 --> 00:26:04.105 all these granular details were moved over or 00:26:04.105 --> 00:26:08.985 these, how do you put encoded behaviors, you 00:26:08.985 --> 00:26:11.970 know, the semantics of the actual code, if 00:26:11.970 --> 00:26:13.890 those weren't brought over as well, then it 00:26:13.890 --> 00:26:16.210 doesn't know it it doesn't know to do 00:26:16.210 --> 00:26:18.929 that. It doesn't know to read into that. 00:26:18.929 --> 00:26:24.210 And the memory being reset or, you know, 00:26:24.210 --> 00:26:28.154 having a limited working memory is often the 00:26:28.154 --> 00:26:33.054 biggest, like, reason why this has to, why 00:26:33.115 --> 00:26:35.274 why why this is a drawback or why 00:26:35.274 --> 00:26:38.575 why it's holding us back. You know? So 00:26:39.049 --> 00:26:41.210 Cloud Code v two point one one six 00:26:41.210 --> 00:26:43.450 introduced a new task system replacing the old 00:26:43.450 --> 00:26:45.850 to do list with something more robust. The 00:26:45.850 --> 00:26:48.809 key difference, tasks support dependency graphs. Task three 00:26:48.809 --> 00:26:50.570 can't start until task one and task two 00:26:50.570 --> 00:26:53.245 complete. Preventing the AI from trying to test 00:26:53.245 --> 00:26:55.485 code it hasn't written yet. Tasks are written 00:26:55.485 --> 00:26:57.405 to your local file system, so they survive 00:26:57.405 --> 00:27:00.125 crashes and can be version controlled. Multiple cloud 00:27:00.125 --> 00:27:02.385 sessions can share the same task list, enabling 00:27:02.445 --> 00:27:05.500 patterns like writer reviewer where one session writes 00:27:05.500 --> 00:27:07.420 code and another reviews it in a clean 00:27:07.420 --> 00:27:10.540 context. Recent patches fixed out of memory crashes 00:27:10.540 --> 00:27:15.020 and added headless mode for CICD integration. And 00:27:15.020 --> 00:27:17.980 headless mode is another fun little tool that 00:27:17.980 --> 00:27:22.225 you can use because, technically, you can run 00:27:22.225 --> 00:27:27.265 Claude as a command line, prompt that which 00:27:27.265 --> 00:27:28.705 which would you don't have to open the 00:27:28.705 --> 00:27:31.365 TUI. You can you can prompt it, headlessly. 00:27:31.985 --> 00:27:34.920 And if you have a script that you, 00:27:34.920 --> 00:27:37.020 you know, maybe say you pull a document 00:27:37.240 --> 00:27:39.800 automated, you have a automated document that you 00:27:39.800 --> 00:27:42.060 pull from online, and you have a script 00:27:42.200 --> 00:27:48.774 that calls your automated, you know, Claude, in 00:27:48.774 --> 00:27:51.335 headless mode, which with it's a prompt you 00:27:51.335 --> 00:27:53.095 already saved in your code, you know, and 00:27:53.095 --> 00:27:56.455 you just have to exec that, and you 00:27:56.615 --> 00:27:59.575 you'll have an output automate automatically. So I 00:27:59.575 --> 00:28:02.235 use this to generate sentiment on certain things, 00:28:04.870 --> 00:28:07.030 which I I find fantastic. You know? It's 00:28:07.030 --> 00:28:09.510 a one liner, very limited amount of token 00:28:09.510 --> 00:28:11.830 use, and I think it's a very clever 00:28:11.830 --> 00:28:18.195 way to, yeah, understand sentiment overall of larger 00:28:18.195 --> 00:28:20.215 pieces of works before you get into it. 00:28:23.075 --> 00:28:26.370 But this is kinda cool. Claude is moving 00:28:26.770 --> 00:28:29.730 copilot to project manager, and I did see 00:28:29.730 --> 00:28:32.210 another tool called Tracer AI, which was more 00:28:32.210 --> 00:28:36.690 of a project manager style system. But the 00:28:36.690 --> 00:28:39.410 the persistence seems to be where we're we're 00:28:39.410 --> 00:28:41.990 losing out, and and they're starting to recognize 00:28:42.255 --> 00:28:46.095 this, you know. Before clearing context to free 00:28:46.095 --> 00:28:48.815 up tokens meant losing the plan, but now 00:28:48.815 --> 00:28:50.655 the plan lives on disk so you can 00:28:50.655 --> 00:28:53.795 aggressively manage context without losing your road map. 00:28:54.095 --> 00:28:56.995 For enterprises, the plan becomes an audible artifact. 00:28:57.400 --> 00:28:59.480 For individuals, you can shut down your terminals, 00:28:59.480 --> 00:29:01.480 switch machines, or recover from a crash, and 00:29:01.480 --> 00:29:07.559 pick up exactly where you left off. This 00:29:07.559 --> 00:29:11.875 is kinda cool. I was already doing something 00:29:11.875 --> 00:29:16.755 similar to this through something I call, well, 00:29:16.755 --> 00:29:19.655 Claude has these command skills that you can 00:29:22.115 --> 00:29:26.490 use to have it run through specific, instructions 00:29:26.490 --> 00:29:29.130 that you set up ahead of time. And 00:29:29.130 --> 00:29:31.850 I have this workflow instruction that allow you 00:29:31.850 --> 00:29:35.470 know, does a little bit of research, documents 00:29:35.530 --> 00:29:38.890 all of its findings, you know, and then 00:29:38.890 --> 00:29:41.815 derives, you know, a plan based on those 00:29:41.815 --> 00:29:44.294 findings. And then, you know, I tend to 00:29:44.294 --> 00:29:47.355 go back and forth on those, plans before 00:29:47.414 --> 00:29:49.975 trying them out, but it's been kind of 00:29:49.975 --> 00:29:52.375 a fun thing to try out with, my 00:29:52.375 --> 00:29:55.040 personal project that I've been building up, over 00:29:55.040 --> 00:30:00.960 the last few months. But, yeah, pretty awesome 00:30:00.960 --> 00:30:03.679 stuff. I think, this this is definitely heading 00:30:03.679 --> 00:30:06.179 in the right direction for the coding agents, 00:30:07.924 --> 00:30:11.125 because the memory is definitely currently the biggest 00:30:11.125 --> 00:30:14.985 limitation. More cloud code stuff. Cloud code teleport. 00:30:15.284 --> 00:30:17.945 Remove your coding sessions between web and terminal 00:30:18.005 --> 00:30:20.485 terminal. You can see where my my articles 00:30:20.485 --> 00:30:21.605 are at. I guess I'm just getting a 00:30:21.605 --> 00:30:23.970 lot of cloud stuff right now. Especially with 00:30:23.970 --> 00:30:25.890 the OpenClaw stuff, I can kinda see why 00:30:25.890 --> 00:30:28.210 this is happening. Cloud Code has kind of 00:30:28.210 --> 00:30:36.485 been the the reigning, you know, code generation 00:30:36.705 --> 00:30:41.425 tool amongst all of the different, available, you 00:30:41.425 --> 00:30:44.065 know, between Gemini and OpenAI and whatnot. It 00:30:44.065 --> 00:30:46.225 seems that Cloud Code's the most popular and 00:30:46.225 --> 00:30:49.929 always has the highest scoring on the, you 00:30:49.929 --> 00:30:52.889 know, the those, like, test ratings. So it's 00:30:52.889 --> 00:30:58.330 kinda interesting more and more to see, more 00:30:58.330 --> 00:31:02.375 Claude articles coming up, especially in my my 00:31:02.375 --> 00:31:06.135 little feed that I use to hunt for 00:31:06.135 --> 00:31:11.255 articles. But, this is from Hacker Noon written 00:31:11.255 --> 00:31:16.990 by Ladislav Guzzi. One frustration with AI coding 00:31:16.990 --> 00:31:20.430 tools has been, again, context switching. Start a 00:31:20.430 --> 00:31:22.190 conversation on the web, realize you need to 00:31:22.190 --> 00:31:25.790 work locally, and you're essentially starting over, or 00:31:25.790 --> 00:31:27.790 you're on your laptop, get interrupted, and wish 00:31:27.790 --> 00:31:30.095 you could pick up on another machine. Cloud 00:31:30.095 --> 00:31:33.135 Code's new teleport feature lets developers transfer entire 00:31:33.135 --> 00:31:36.895 coding sessions, including conversation history context and the 00:31:36.895 --> 00:31:41.295 working branch between web and local environments. Using 00:31:41.295 --> 00:31:44.255 an ampersand prefix sends a task to Cloud 00:31:44.255 --> 00:31:47.870 Code's cloud where it continues running even after 00:31:47.870 --> 00:31:50.510 you close your editor. The teleport command pulls 00:31:50.510 --> 00:31:52.910 web sessions back to your local terminal. You 00:31:52.910 --> 00:31:55.330 can share session IDs with teammates for asynchronous 00:31:55.470 --> 00:31:59.185 pair programming. The catch, it's one way only. 00:31:59.405 --> 00:32:01.485 You can pull web sessions locally, but can't 00:32:01.485 --> 00:32:05.085 push existing local sessions to the web. And 00:32:05.085 --> 00:32:07.485 it looks like this requires GitHub integration and 00:32:07.485 --> 00:32:09.660 a clean git skate git Git Slate. You 00:32:09.740 --> 00:32:11.580 know, it's funny because this sounds like something 00:32:11.580 --> 00:32:17.820 that OpenClaw would already do. So now it 00:32:17.820 --> 00:32:20.620 seems like it's it's tough because so Anthropic 00:32:20.620 --> 00:32:22.780 has to build not only the tool that 00:32:22.780 --> 00:32:25.100 drives it, but also they have to improve 00:32:25.100 --> 00:32:27.075 their wrapper, right, which in this case is 00:32:27.075 --> 00:32:31.895 Claude Cook. But OpenClaus is gets to benefit 00:32:31.955 --> 00:32:34.355 off of all their improvements and just has 00:32:34.355 --> 00:32:38.275 to improve their wrapper. So you can see 00:32:38.275 --> 00:32:44.700 where the, requirement difference is in terms of, 00:32:44.700 --> 00:32:47.020 you know, these businesses and how they're meeting 00:32:47.020 --> 00:32:51.580 people's needs. But if you don't wanna use 00:32:51.580 --> 00:32:54.400 open clause, definitely solves a real workflow problem 00:32:54.780 --> 00:32:57.665 for people. You can start complex work on 00:32:57.665 --> 00:32:59.425 the web during the commute and teleport it 00:32:59.425 --> 00:33:01.825 to your workstation when you arrive, hand off 00:33:01.825 --> 00:33:03.425 a session to a colleague in a different 00:33:03.425 --> 00:33:05.665 time zone, like compute heavy work running the 00:33:05.665 --> 00:33:08.200 cloud while you're working on things locally. The 00:33:08.200 --> 00:33:11.720 one way limitation is notable. And try Anthropic 00:33:11.720 --> 00:33:14.200 likely wants to ensure code integrity through through 00:33:14.200 --> 00:33:18.760 Git, but it's still, like, meaningful towards, like, 00:33:18.760 --> 00:33:22.205 a portable AI assisted development. I still think 00:33:22.205 --> 00:33:24.285 OpenCloud might be a better solution this than 00:33:24.285 --> 00:33:30.545 this, though. Alright. More AI. Google expands Gemini 00:33:30.605 --> 00:33:33.485 API with a hundred megabyte file limits and 00:33:33.485 --> 00:33:39.010 cloud storage integration. This is from Chrome Unboxed 00:33:39.070 --> 00:33:43.230 written by Robbie Payne. Building AI applications that 00:33:43.230 --> 00:33:46.270 handle documents, images, and videos has meant dealing 00:33:46.270 --> 00:33:49.010 with file size restrictions. And, you know, developers 00:33:49.070 --> 00:33:52.605 often had to preprocess content, split large files, 00:33:52.985 --> 00:33:55.545 or build complex workarounds just to feed data 00:33:55.545 --> 00:33:59.385 to AI models. But Google increased the Gemini 00:33:59.385 --> 00:34:02.985 API's inline file size from twenty megabytes to 00:34:02.985 --> 00:34:06.750 a hundred megabytes, a fivefold increase allowing developers 00:34:06.810 --> 00:34:10.090 to send high resolution images, extended audio clips, 00:34:10.090 --> 00:34:15.050 and complex PDFs directly without intermediate storage. New 00:34:15.050 --> 00:34:17.690 integration with Google Cloud Storage lets developers register 00:34:17.690 --> 00:34:20.845 files once and reference them across multiple requests. 00:34:21.625 --> 00:34:24.665 The API also now supports fetching content via 00:34:24.665 --> 00:34:28.744 HTTPS and signed URLs from competing platforms like 00:34:28.744 --> 00:34:32.799 AWS s three or Azure Blob Storage. These 00:34:32.799 --> 00:34:35.520 capabilities run on Gemini three models designed for 00:34:35.520 --> 00:34:40.640 complex reasoning across large multimodal inputs. So I 00:34:40.640 --> 00:34:42.480 guess if any of you are developers building 00:34:42.480 --> 00:34:45.119 AI applications, this might remove some of your 00:34:45.119 --> 00:34:48.325 friction. You now feed an AI a forty 00:34:48.325 --> 00:34:50.404 five minute video or a hundred paid document 00:34:50.404 --> 00:34:55.865 directly. And maybe, I guess, it just runs 00:34:56.244 --> 00:35:03.420 on that. I guess, maybe this might be 00:35:03.420 --> 00:35:06.060 a little more valuable if you're uploading a 00:35:06.060 --> 00:35:10.480 lot of, I don't know, high level documentation 00:35:11.020 --> 00:35:13.600 with tons of detail. I I really wonder, 00:35:13.660 --> 00:35:19.775 you know, But this could also be real 00:35:20.015 --> 00:35:23.295 really good for maybe, like, complex reasoning across, 00:35:23.295 --> 00:35:27.135 like, large reports or many, many reports so 00:35:27.135 --> 00:35:30.000 you could, you know, put more a larger 00:35:30.000 --> 00:35:33.060 report or a larger time frame report in 00:35:33.200 --> 00:35:38.740 to be analyzed rather than twenty megabytes worth. 00:35:39.360 --> 00:35:41.385 So I don't know. This sounds like a 00:35:41.385 --> 00:35:44.125 nice bonus for anyone that uses Gemini API. 00:35:45.705 --> 00:35:49.785 Another Gemini fun bit, from the Google AI 00:35:49.785 --> 00:35:57.330 blog, written by Rohan Doshi. Google launches agentic 00:35:57.330 --> 00:35:59.810 vision in Gemini three flash. Oh, yeah. Yeah. 00:35:59.810 --> 00:36:02.530 This this one is kind of cool. When 00:36:02.530 --> 00:36:04.450 AI looks at an image today, it typically 00:36:04.450 --> 00:36:06.690 describes what it sees in a single pass. 00:36:06.690 --> 00:36:09.090 It's like glancing at a photo versus actually 00:36:09.090 --> 00:36:11.905 studying it. But some tasks, like reading dense 00:36:11.905 --> 00:36:15.845 tables, checking architectural plans, or analyzing detailed diagrams, 00:36:16.145 --> 00:36:19.825 need more than a quick look. Google's new 00:36:19.825 --> 00:36:23.365 agentic vision capability capability in Gemini three Flash 00:36:23.570 --> 00:36:27.170 treats image understanding as an active investigation rather 00:36:27.170 --> 00:36:30.450 than a single observation. The model can formulate 00:36:30.450 --> 00:36:33.830 plans to zoom in, inspect, and manipulate images 00:36:33.970 --> 00:36:36.690 step by step using Python code execution to 00:36:36.690 --> 00:36:40.755 draw on images, parse tables, and visualize data. 00:36:42.015 --> 00:36:45.215 This think, act, observe loop lets AI ground 00:36:45.215 --> 00:36:48.435 its answers in visual evidence rather than guessing. 00:36:48.975 --> 00:36:51.535 Testing shows a consistent five to ten percent 00:36:51.535 --> 00:36:55.050 quality improvement across vision benchmarks. One company using 00:36:55.050 --> 00:36:58.330 it for building plan validation improved accuracy by 00:36:58.330 --> 00:37:02.410 five percent just by enabling code execution, and 00:37:02.410 --> 00:37:06.345 that's for the, Python scripts there. And this 00:37:06.345 --> 00:37:08.985 is AI that can actually study an image 00:37:08.985 --> 00:37:12.025 rather than describe it. It's it's let you 00:37:12.025 --> 00:37:15.325 know, this this takes away the the opportunity 00:37:15.865 --> 00:37:18.744 for it to hallucinate. It builds more context 00:37:18.744 --> 00:37:20.585 grounded in the image it's looking at. You 00:37:20.585 --> 00:37:22.460 know what I mean? So if if you 00:37:22.460 --> 00:37:24.559 ask it to count items in a complex 00:37:24.619 --> 00:37:26.700 photo, it can write code to mark each 00:37:26.700 --> 00:37:29.740 one as it counts. If you need data 00:37:29.740 --> 00:37:31.980 extracted from a messy table, you can use 00:37:31.980 --> 00:37:35.040 Python to process it systematically rather than guessing. 00:37:35.565 --> 00:37:38.685 The practical impact, fewer hallucinations when AI deals 00:37:38.685 --> 00:37:41.885 with visual information that requires careful attention to 00:37:41.885 --> 00:37:50.480 detail. On to the next article. Researchers claim 00:37:50.480 --> 00:37:55.120 LLMs will never truly be intelligent. Tech leaders 00:37:55.120 --> 00:37:56.960 keep promising that if we just scale language 00:37:56.960 --> 00:38:00.080 models big enough, we'll we'll achieve artificial generated 00:38:00.080 --> 00:38:03.155 intelligence. Billions of dollars are being invested on 00:38:03.155 --> 00:38:05.875 this premise, but not everyone's convinced. So this 00:38:05.875 --> 00:38:08.115 article, just to be clear, is not yet, 00:38:08.355 --> 00:38:11.895 peer reviewed, but I think it gives a 00:38:13.839 --> 00:38:17.839 interesting context or interesting perspective into why, you 00:38:17.839 --> 00:38:20.320 know, the the research being done to to 00:38:20.320 --> 00:38:24.480 prove that, you know, LLMs are not they 00:38:24.480 --> 00:38:26.560 may seem close to what we what we're 00:38:26.560 --> 00:38:30.765 trying to achieve, but but this researcher is 00:38:30.765 --> 00:38:33.565 convinced that it is not gonna lead to 00:38:33.565 --> 00:38:38.365 AGI. So Benjamin Riley is the name of 00:38:38.365 --> 00:38:40.765 the researcher, but this article from futurism is 00:38:40.765 --> 00:38:45.010 written by Frank Landy Moore. But Benjamin Riley, 00:38:45.010 --> 00:38:48.609 cofounder of Cognitive Resonance, argues that language and 00:38:48.609 --> 00:38:53.089 intelligence are fundamentally separate cognitive functions. His core 00:38:53.089 --> 00:38:56.849 claim, human thinking is largely independent of human 00:38:56.849 --> 00:38:59.655 language. He points to neuroscience research showing that 00:38:59.655 --> 00:39:02.455 people who lose language abilities retain the cap 00:39:02.695 --> 00:39:06.855 capacity to solve math problems, understand emotions, and 00:39:06.855 --> 00:39:10.455 reason about the world. LLMs, he argues, merely 00:39:10.455 --> 00:39:15.430 simulate communicate communicative patterns rather than actually thinking 00:39:15.650 --> 00:39:19.490 correct. Research on AI creativity suggests these systems 00:39:19.490 --> 00:39:23.830 hit inherent ceilings, producing formulaic rather than truly 00:39:23.890 --> 00:39:27.545 novel outputs. Even Yan LeCun, one of the 00:39:27.545 --> 00:39:30.665 pioneers of modern AI, disputes that scaling the 00:39:30.665 --> 00:39:35.885 LLMs will lead to AGI, advocating instead for 00:39:35.944 --> 00:39:41.250 alternative world model approaches. I am on this 00:39:41.250 --> 00:39:44.950 boat as well. At this point in time, 00:39:45.250 --> 00:39:48.630 I do believe that we are merely being 00:39:48.930 --> 00:39:51.330 fooled into believing that we are seeing some 00:39:51.330 --> 00:39:55.875 sort of intelligence. But, again, the data proves 00:39:55.934 --> 00:40:02.355 that this is merely pattern matching system, and 00:40:02.414 --> 00:40:05.934 we need to keep that in our minds 00:40:05.934 --> 00:40:08.174 as we use these tools that, you know, 00:40:08.174 --> 00:40:13.990 that we do not personify them. So I 00:40:13.990 --> 00:40:15.690 feel like this is a pretty good counterweight 00:40:15.830 --> 00:40:17.670 to a lot of the AGI hype. I'm 00:40:17.670 --> 00:40:21.830 hoping that, you know, that this idea or 00:40:21.830 --> 00:40:26.635 belief picks up more steam and and becomes 00:40:26.635 --> 00:40:29.035 grounded in more empirical data. That's that's that's 00:40:29.035 --> 00:40:30.795 the important part, you know, rather than it 00:40:30.795 --> 00:40:33.915 being a belief that it's a fact. And 00:40:33.915 --> 00:40:36.474 the question isn't whether LMS are useful. They 00:40:36.474 --> 00:40:39.030 they are. The question is whether making them 00:40:39.030 --> 00:40:42.390 bigger will eventually produce genuine intelligence. And you 00:40:42.390 --> 00:40:44.310 think about the amount of scaling we're doing 00:40:44.310 --> 00:40:46.550 in some of these states in terms of, 00:40:46.950 --> 00:40:53.445 megawatts. So I don't know. Riley's argument suggests 00:40:53.445 --> 00:40:56.505 that we might be optimizing the wrong thing. 00:40:56.645 --> 00:40:59.525 Language is a communication tool, not a thinking 00:40:59.525 --> 00:41:02.185 tool. That doesn't mean AI progress will stop, 00:41:02.250 --> 00:41:03.850 but it might mean the path to more 00:41:03.850 --> 00:41:08.090 capable AI runs through fundamentally different architectures, not 00:41:08.090 --> 00:41:12.090 just bigger language models. I would be interested 00:41:12.090 --> 00:41:21.545 to see where this goes. Let's move on. 00:41:21.545 --> 00:41:24.424 I think, I got another few for you 00:41:24.424 --> 00:41:26.664 folks here. I actually pretty much loaded up 00:41:26.664 --> 00:41:29.085 today on on a ton of articles. So, 00:41:29.619 --> 00:41:31.940 you can see I've been stockpiling for a 00:41:31.940 --> 00:41:35.300 little while, I guess, but paper claims AI 00:41:35.300 --> 00:41:40.740 agents are mathematically limited in complex tasks. So 00:41:40.740 --> 00:41:47.135 this is another, AI limitational of paper I 00:41:47.135 --> 00:41:50.495 found on futurism, written, again, by Frankie Landy 00:41:50.495 --> 00:41:52.575 Moore or Frank Landy Moore. So you you 00:41:52.815 --> 00:41:54.655 I think Frank's on the same side as 00:41:54.655 --> 00:41:57.455 me. We don't think that that that is 00:41:57.455 --> 00:41:59.695 coming to AGI since they they've got two 00:41:59.695 --> 00:42:03.430 papers now. But AI agents systems that can 00:42:03.430 --> 00:42:05.510 take actions rather than just generate text are 00:42:05.510 --> 00:42:08.230 being deployed for everything from customer service to 00:42:08.230 --> 00:42:11.030 software development. Companies are betting they can replace 00:42:11.030 --> 00:42:14.470 human workers. But, you know, how reliable are 00:42:14.470 --> 00:42:17.635 they? A non peer reviewed paper by Vishal 00:42:18.335 --> 00:42:21.315 Sikka and his son claims to be mathematically 00:42:22.815 --> 00:42:26.575 claims to mathematically demonstrate that large language models 00:42:26.575 --> 00:42:30.520 are incapable of carrying out computational and agentic 00:42:30.579 --> 00:42:33.780 tasks beyond a certain complexity, and that threshold 00:42:33.780 --> 00:42:35.859 is surprisingly low. Okay. So it seems like 00:42:35.859 --> 00:42:39.700 I confused the two articles because this is 00:42:39.700 --> 00:42:42.660 the one with the non peer reviewed, paper 00:42:42.660 --> 00:42:45.300 rather than the previous one by Frank Landy 00:42:45.300 --> 00:42:49.275 Moore. Whoops. Sorry about that, guys. But the 00:42:49.275 --> 00:42:52.495 paper points to real world failures. Companies deploying 00:42:52.635 --> 00:42:55.435 AI agents for workforce reduction discovered the system 00:42:55.435 --> 00:42:59.195 couldn't adequately replace human workers due to frequent 00:42:59.195 --> 00:43:02.779 errors. OpenAI research has conceded that AI hallucinations 00:43:03.079 --> 00:43:06.619 remain pervasive and that model accuracy will never 00:43:06.680 --> 00:43:10.039 reach a hundred percent. Sitka argues AI agents 00:43:10.039 --> 00:43:13.645 cannot safely operate in high stakes environments, Those 00:43:13.645 --> 00:43:17.405 supplementary components built around LLMs might overcome pure 00:43:17.405 --> 00:43:22.205 model limitations. So but, yeah, take this with 00:43:22.205 --> 00:43:25.005 a grain of salt. The claims are pretty 00:43:25.005 --> 00:43:26.829 strong in this paper, but it it aligns 00:43:26.829 --> 00:43:28.670 with what many people are seeing, like, in 00:43:28.670 --> 00:43:30.910 practice that AI agents work great for some 00:43:30.910 --> 00:43:33.710 task and and just, you know, they just 00:43:33.710 --> 00:43:38.450 fail spectacularly at others, with limited predictability about 00:43:38.510 --> 00:43:40.530 which is which. You know, you you really 00:43:40.829 --> 00:43:44.414 don't know Well, you can kinda you you 00:43:44.414 --> 00:43:47.615 kinda do know. I I still am of 00:43:47.615 --> 00:43:51.454 the belief that, you know, that to a 00:43:51.454 --> 00:43:54.255 degree when you give it a large, broad 00:43:54.255 --> 00:43:59.260 strokes task, that the the granular details should 00:43:59.260 --> 00:44:01.980 be reviewed and re encoded in because that 00:44:01.980 --> 00:44:04.799 seems to be, like, something that's consistently missed. 00:44:07.740 --> 00:44:11.099 But the company is rushing to replace workers 00:44:11.099 --> 00:44:12.940 with AI agents may be getting ahead of 00:44:12.940 --> 00:44:16.555 technology's actual reliability. And never reaching a hundred 00:44:16.555 --> 00:44:19.194 percent point from open AI researchers is pretty 00:44:19.194 --> 00:44:22.474 notable. If even the makers at Minted Tools 00:44:22.474 --> 00:44:25.035 will always have error rates, that limits where 00:44:25.035 --> 00:44:29.330 they can safely be deployed. So this is 00:44:29.330 --> 00:44:31.990 pretty interesting. I wouldn't mark it as settled 00:44:32.130 --> 00:44:37.570 just yet, but, definitely, the common sentiment that's 00:44:37.570 --> 00:44:39.570 being passed around in general, I feel like, 00:44:39.570 --> 00:44:48.065 matches what this article is saying. But okay. 00:44:48.065 --> 00:44:50.385 I'm gonna hop. I'm just gonna skip forward 00:44:50.385 --> 00:44:52.545 to a non AI article so I can 00:44:52.545 --> 00:44:56.325 free you folks from my absolute AI onslaught. 00:44:56.385 --> 00:44:58.625 I didn't realize I got so many AI 00:44:58.625 --> 00:45:03.570 articles. This one's from Ars Technica. Haven one, 00:45:03.570 --> 00:45:07.250 first commercial space station now being assembled for 00:45:07.250 --> 00:45:12.690 launch. With NASA planning to retire the of 00:45:12.690 --> 00:45:14.555 the ISS, which has been a very common 00:45:14.555 --> 00:45:17.035 topic in my shows, there's a race to 00:45:17.035 --> 00:45:19.515 build commercial replacements. And we've talked about this, 00:45:19.515 --> 00:45:21.835 that that that in twenty thirty, we're gonna 00:45:21.835 --> 00:45:24.954 start seeing the research be commercialized rather than 00:45:24.954 --> 00:45:30.559 it being handled by NASA. And a large 00:45:30.559 --> 00:45:32.960 concern with this is the the large share 00:45:32.960 --> 00:45:36.160 that, you know, SpaceX has over the current 00:45:36.160 --> 00:45:39.359 market in terms of space development and space 00:45:39.359 --> 00:45:41.394 research. It it you know, last we checked, 00:45:41.394 --> 00:45:43.234 it was, like, seventy five percent or something 00:45:43.234 --> 00:45:45.555 of the contracts are in the hands of 00:45:45.555 --> 00:45:50.595 SpaceX. So ISS has been humanity's home in 00:45:50.595 --> 00:45:52.674 orbit for over two decades, but it's aging 00:45:52.674 --> 00:45:55.720 and expensive to maintain. Several companies are working 00:45:55.720 --> 00:45:58.940 on private station, but VAST is pulling ahead. 00:45:59.640 --> 00:46:03.640 VAST CEO Max Hout confirmed to Ars Technica 00:46:03.640 --> 00:46:05.720 that Haven one's launch has slipped from mid 00:46:05.720 --> 00:46:07.960 twenty twenty six to q one twenty twenty 00:46:07.960 --> 00:46:10.945 seven, their first public acknowledgment of the delay. 00:46:11.245 --> 00:46:13.965 The good news, the fifteen ton station's primary 00:46:13.965 --> 00:46:16.685 structure was fully completed on January tenth, and 00:46:16.685 --> 00:46:19.245 they are now starting clean room integration of 00:46:19.245 --> 00:46:24.279 thermal control, propulsion, interior shells, and avionics. Final 00:46:24.279 --> 00:46:27.000 closeout expected by fall followed by a full 00:46:27.000 --> 00:46:31.640 test campaign at NASA's Plum Brook facility. Once 00:46:31.640 --> 00:46:34.519 launched at Falcon nine without crew, the station 00:46:34.519 --> 00:46:37.695 could receive its first visitors as early as 00:46:37.695 --> 00:46:40.575 two weeks later once SpaceX certifies Dragon can 00:46:40.575 --> 00:46:44.095 safely dock. VAST is in deep negotiations with 00:46:44.095 --> 00:46:47.615 private individuals and nation states for crew and 00:46:47.615 --> 00:46:49.775 believe they can compress training to six months 00:46:49.775 --> 00:46:54.720 for experienced astronauts. This is still a pretty 00:46:54.720 --> 00:46:59.860 genuine milestone. First privately built commercial space station, 00:47:00.000 --> 00:47:03.280 likely two years ahead of any competitor. How 00:47:03.280 --> 00:47:05.920 it has made a pointed comment about NASA's 00:47:05.920 --> 00:47:10.055 delayed requirements. We think they should focus on 00:47:10.055 --> 00:47:12.454 what is right for the country, not on 00:47:12.454 --> 00:47:16.795 being fair to all bidders. With IRS's retirement 00:47:16.855 --> 00:47:19.015 less than five years away, he's urging NASA 00:47:19.015 --> 00:47:22.590 to require a crew demonstration flight soon, not 00:47:22.590 --> 00:47:24.990 on the eve of twenty thirty one. The 00:47:24.990 --> 00:47:26.990 station will have a three year lifetime with 00:47:26.990 --> 00:47:30.750 up to four crews visited up to four 00:47:30.750 --> 00:47:36.845 crew visits planned, ten day missions each. Wow. 00:47:39.625 --> 00:47:41.224 That's a pretty cool one. So, yeah, we're 00:47:41.224 --> 00:47:43.545 starting to see the first ones, kinda come 00:47:43.545 --> 00:47:46.365 out to market. I guess not market, but 00:47:46.425 --> 00:47:51.840 come out to public. And those are definitely 00:47:51.980 --> 00:47:57.900 looking, pretty sweet. This one's from, space dot 00:47:57.900 --> 00:48:01.180 com. I got a article by by Josh 00:48:01.180 --> 00:48:06.055 Dinner. Spaceflight shifts astronauts' brains inside their skulls. 00:48:07.155 --> 00:48:09.475 As space agencies plan longer missions to the 00:48:09.475 --> 00:48:12.595 moon, to Mars, understanding what extended spaceflight does 00:48:12.595 --> 00:48:14.910 to the human body becomes critical. We know 00:48:14.910 --> 00:48:18.030 about muscle loss and bone density issues. The 00:48:18.030 --> 00:48:20.349 effects on the brain though are are less 00:48:20.349 --> 00:48:23.970 well understood. So we got some new research, 00:48:24.510 --> 00:48:30.545 regarding this. But, using neuroimaging, they, they've documented 00:48:30.925 --> 00:48:34.845 that microgravity causes astronauts' brains to physically shift 00:48:34.845 --> 00:48:39.105 position within their skulls, specifically backward and upward 00:48:39.165 --> 00:48:43.020 and tilted. Scientists trained brain or tracked brain 00:48:43.020 --> 00:48:46.640 positioning changes across missions of varying durations, approximately 00:48:46.700 --> 00:48:49.580 two weeks, six months, and one year. Their 00:48:49.580 --> 00:48:53.260 displacement represents a measurable anatomical change as the 00:48:53.260 --> 00:48:59.414 body adapts to weightlessness. It sounds pretty alarming, 00:48:59.414 --> 00:49:03.815 but I think, it's just kinda interesting from 00:49:03.815 --> 00:49:05.974 a science perspective. It's not like your brain 00:49:05.974 --> 00:49:07.974 was ever fixed in place, you know, the 00:49:07.974 --> 00:49:11.130 first place. It's been floating in our fluid 00:49:11.130 --> 00:49:13.369 brains, and gravity kind of affects where it's 00:49:13.369 --> 00:49:15.050 six. So if you take away the gravity 00:49:15.050 --> 00:49:18.650 and it repositions, the question people are trying 00:49:18.650 --> 00:49:20.330 to answer is where what's going on with 00:49:20.330 --> 00:49:24.974 that? And even more so is whether this 00:49:24.974 --> 00:49:28.174 causes any functional problems, especially on missions lasting 00:49:28.174 --> 00:49:31.055 months or years. As we plan trips to 00:49:31.055 --> 00:49:33.055 Mars that could take three year round trips, 00:49:33.055 --> 00:49:35.375 understanding these effects is pretty essential. We're keeping 00:49:35.375 --> 00:49:43.330 our astronauts alive and healthy. I have one 00:49:43.330 --> 00:49:47.410 final one for you, folks, this afternoon. This 00:49:47.410 --> 00:49:53.355 one's from, the Veterans Associations, actually. So IBM 00:49:53.355 --> 00:49:56.075 skills build free tech training for veterans and 00:49:56.075 --> 00:50:00.715 military families. If you're transitioning from military service 00:50:00.715 --> 00:50:02.875 to civilian careers, you know it can be 00:50:02.875 --> 00:50:06.859 pretty challenging. We've all heard, the difficulties that 00:50:06.859 --> 00:50:09.619 a lot of veterans face, transitioning to civilian 00:50:09.619 --> 00:50:14.420 life, especially in tech fields where credentials and 00:50:14.420 --> 00:50:18.020 specific skills matter. I remember I was listening 00:50:18.020 --> 00:50:21.565 to one of my coworker shows, and he 00:50:21.565 --> 00:50:26.685 was talking about how his avionics, certification didn't 00:50:26.685 --> 00:50:29.905 transfer over when he left the military. And 00:50:30.285 --> 00:50:32.785 it required a whole different set of documentation, 00:50:32.925 --> 00:50:34.940 which I found to be insane. You know, 00:50:34.940 --> 00:50:37.039 you're you're trained on the exact same stuff. 00:50:41.020 --> 00:50:43.660 But the VA has been partnering with industry 00:50:43.660 --> 00:50:46.619 to kinda create pathways for these veterans to, 00:50:46.619 --> 00:50:51.005 you know, find work. So IBM SkillsBuild is 00:50:51.005 --> 00:50:54.205 a free program offering over one thousand online 00:50:54.205 --> 00:50:57.585 courses in high demand fields, including data analytics, 00:50:58.045 --> 00:51:03.140 cybersecurity, AI, cloud computing, and IT support. It's 00:51:03.140 --> 00:51:06.260 available to veterans, active duty service members, national 00:51:06.260 --> 00:51:09.780 guard, and reserve members, their spouses, children, and 00:51:09.780 --> 00:51:13.860 caregivers, anyone eighteen or older. The program offers 00:51:13.860 --> 00:51:17.065 two formats, structured live courses with weekly plans 00:51:17.065 --> 00:51:19.965 and support, and self paced twenty four access 00:51:20.825 --> 00:51:25.145 courses. Complete this earn industry recognized credentials that 00:51:25.145 --> 00:51:27.945 can be added to LinkedIn profiles. That's actually 00:51:27.945 --> 00:51:30.050 kinda sick. If If you're a veteran or 00:51:30.050 --> 00:51:31.570 have veterans in your life looking to move 00:51:31.570 --> 00:51:33.730 into tech, this is a pretty legit, free 00:51:33.730 --> 00:51:38.050 resource worth knowing about. IBM credentials carry weight 00:51:38.050 --> 00:51:40.290 with a lot of employers, and the program 00:51:40.290 --> 00:51:43.275 covers areas where jobs are plentiful. The inclusion 00:51:43.275 --> 00:51:45.434 of family members is also pretty notable. It's 00:51:45.434 --> 00:51:47.675 not just for your serve service members, but 00:51:47.675 --> 00:51:50.494 your your your wife, your children as well. 00:51:50.635 --> 00:51:52.954 Sign up at the IBM skill build skills 00:51:52.954 --> 00:51:56.315 build website, and you'll automatically get directed to 00:51:56.315 --> 00:52:03.160 the veteran specific section. Before we close out, 00:52:03.160 --> 00:52:04.839 you know, and, you know, something to think 00:52:04.839 --> 00:52:07.640 about coming into this next week, but mold 00:52:07.640 --> 00:52:10.200 book, AI to AI communication, what do you 00:52:10.200 --> 00:52:13.484 guys think about this? About these AI agents 00:52:13.484 --> 00:52:19.005 forming communities, debating philosophy, that new Christopher because 00:52:19.165 --> 00:52:22.605 what did they call it? Christopher I can't 00:52:22.605 --> 00:52:28.720 even pronounce it. The religion is called crustopher 00:52:29.340 --> 00:52:31.900 crustopherianism. So, I mean, you know, look into 00:52:31.900 --> 00:52:33.600 that. What do you guys think of crustopherianism? 00:52:35.900 --> 00:52:37.980 Are you planning to try the open claw 00:52:37.980 --> 00:52:41.595 system? It definitely sounds like it's going to 00:52:41.595 --> 00:52:47.695 be the new wave and and and how 00:52:48.395 --> 00:52:53.515 we're going to be moving forward into using 00:52:53.515 --> 00:52:57.230 these systems. Everything is now becoming about the 00:52:57.230 --> 00:53:00.190 context, and I'm sure we're gonna continue to 00:53:00.190 --> 00:53:02.670 see the price of memory go up. For 00:53:02.910 --> 00:53:05.150 speaking of which, I did end up buying 00:53:05.150 --> 00:53:08.625 my new PC, but I still got wrecked 00:53:08.625 --> 00:53:14.625 on those memory prices, man. And, it's I 00:53:14.625 --> 00:53:17.825 don't know. Stuff like that is pretty tragic, 00:53:17.825 --> 00:53:20.065 I feel like, to see that overflow into 00:53:20.065 --> 00:53:22.225 other sectors. I mean, it's it's still tech, 00:53:22.225 --> 00:53:25.090 I guess, but but and the and the 00:53:25.090 --> 00:53:26.610 other thing I was also wanting to mention 00:53:26.610 --> 00:53:28.550 was, you know, the whole vibe coding paradox 00:53:28.850 --> 00:53:31.430 where we're we're starting to see the malware 00:53:31.570 --> 00:53:35.330 created by vibe coding, but but vibe coding 00:53:35.330 --> 00:53:38.525 isn't actually that good yet or isn't actually 00:53:38.525 --> 00:53:41.665 that, you know, reliable yet. So it's interesting 00:53:41.725 --> 00:53:46.125 to see these, people come in and try 00:53:46.125 --> 00:53:49.905 to create malware for systems they don't understand. 00:53:50.685 --> 00:53:54.970 And this has actually been a pretty big 00:53:54.970 --> 00:53:56.650 and up and coming thing. And, actually, if 00:53:56.650 --> 00:53:58.329 you guys you guys know I keep up 00:53:58.329 --> 00:54:01.230 with Curl and Daniel Stenberg, he just actually 00:54:01.289 --> 00:54:04.490 closed his bug bounty program due to the 00:54:04.490 --> 00:54:07.529 onslaught of AI slop that he's been getting 00:54:07.529 --> 00:54:10.724 for, you know, bug bounty reports. And I 00:54:10.724 --> 00:54:13.065 I can say we've been getting the same 00:54:13.125 --> 00:54:15.685 freaking slop at at at a blueberry too, 00:54:15.685 --> 00:54:20.484 and it's it's ridiculous, you know, to a 00:54:20.484 --> 00:54:26.220 degree. Overhyping the the degree of the vulnerability, 00:54:26.220 --> 00:54:29.520 you know, not even proving the vulnerability exists, 00:54:29.900 --> 00:54:33.359 creating fake vulnerabilities. You name it. You know? 00:54:33.500 --> 00:54:35.900 They they create these extensive reports that look 00:54:35.900 --> 00:54:38.865 legit and just waste people's time. It just 00:54:38.865 --> 00:54:42.465 wastes people's time. So if you're one of 00:54:42.465 --> 00:54:47.345 those people, please stop. Please stop. I know 00:54:47.345 --> 00:54:49.265 that you might get some free bug bounty 00:54:49.265 --> 00:54:53.250 money, but please stop. You're not helping anyone. 00:54:54.269 --> 00:54:58.190 Well, sorry again for the delayed, show. Gonna 00:54:58.190 --> 00:55:00.190 definitely try to be keeping up this, this 00:55:00.190 --> 00:55:02.269 month and try to put a little bit 00:55:02.269 --> 00:55:04.029 more extra shows in on the on the 00:55:04.029 --> 00:55:07.125 line for you guys. So, definitely keep an 00:55:07.125 --> 00:55:11.045 eye out for that. Happy February first. Stay 00:55:11.045 --> 00:55:13.525 safe out there. Stay warm. If you're out 00:55:13.525 --> 00:55:18.405 in the, the Midwest, definitely bundle up, and